2399

Predicting cutthroat ( clarkii) abundance in high-elevation streams: revisiting a model of translocation success

Michael K. Young, Paula M. Guenther-Gloss, and Ashley D. Ficke

Abstract: Assessing viability of stream populations of (Oncorhynchus clarkii) and identifying streams suitable for establishing populations are priorities in the western United States, and a model was recently developed to predict translocation success (as defined by an index of population size) of two based on mean July water temperature, pool bankfull width, and deep pools counts. To determine whether the translocation model applied to streams elsewhere with more precise abundance estimates, we examined the relation between electrofishing-based esti- mates of cutthroat trout abundance and these habitat variables plus occupied stream length. The preferred model was (population size)1/2 = 0.00508(stream length (m)) + 5.148 (N = 31). In contrast, a model based on data from the origi- nal translocation model included stream temperature and deep pool counts as variables. Differences in models appear to largely have a methodological rather than biological basis. Additional habitat coupled with increased habitat complexity may account for the form of the abundance – stream length relation in the electrofishing-based model. Model-derived estimates imply that many cutthroat trout populations are below thresholds associated with reduced risk of . We believe that this model can reduce uncertainty about projected population sizes when selecting streams for reintroductions or evaluating unsampled streams. Résumé : La détermination de la viabilité des populations de truites fardées (Oncorhynchus clarkii) des cours d’eau et l’identification des milieux d’eau courante adéquats pour l’établissement des populations constituent des priorités dans l’ouest des États-Unis; un modèle mis au point récemment permet de prédire le succès des transferts (défini par un in- dice de taille de population) de deux sous-espèces à partir de la température de l’eau en juillet, de la largeur des fosses à plein bord et du nombre de fosses profondes. Afin de déterminer si le modèle de transfert s’applique à des cours d’eau situés ailleurs avec de meilleures estimations de densité, nous avons examiné la relation entre les estimations d’abondance des truites fardées basées sur la pêche électrique et les variables de l’habitat ci-haut, plus la longueur du cours d’eau occupé. Le meilleur modèle est (taille de la population)1/2 = 0,00508 (longueur du cours d’eau, en m) + 5,148 (N = 31). En revanche, un modèle basé sur les données utilisées dans le modèle original de transfert comprend comme variables la température de l’eau et le nombre de fosses profondes. Les différences entre les modèles semblent avoir une base plus méthodologique que biologique. Un nombre plus important d’habitats et une complexité plus grande des habitats peuvent peut-être expliquer la forme de la relation entre l’abondance et la longueur du cours d’eau dans le modèle basé sur la pêche électrique. Les estimations fournies par les modèles laissent croire que plusieurs des populations de truites fardées se trouvent à une densité inférieure au seuil associé à un risque réduit d’extinction. Nous croyons que ce modèle peut réduire l’incertitude reliée aux projections de tailles de populations lors du choix de cours d’eau pour fins de réintroduction ou pour l’évaluation de cours d’eau non inventoriés. [Traduit par la Rédaction] Young et al. 2408

Introduction Duff 1996). Consequently, biologists have made extensive efforts to found new populations, reestablish extirpated Inland subspecies of cutthroat trout (Oncorhynchus populations, and assess the size and probability of persis- clarkii) have substantially declined throughout most of tence of both (Hepworth et al. 1997; US Fish and Wildlife their historical range over the last 150 years (Behnke 1992; Service 1998; Kruse et al. 2001). Some of this work has

Received 30 July 2004. Accepted 29 June 2005. Published on the NRC Research Press Web site at http://cjm.nrc.ca on 24 September 2005. J18240 M.K. Young.1 US Forest Service, Rocky Mountain Research Station, 800 East Beckwith Avenue, Missoula, MT 59801, USA. P.M. Guenther-Gloss.2 US Forest Service, Medicine Bow-Routt National Forests, Brush Creek – Hayden Ranger District, P.O. Box 249, Saratoga, WY 82331, USA. A.D. Ficke. Department of Wildlife and Fishery Biology, State University, Fort Collins, CO 80526, USA. 1Corresponding author (e-mail: [email protected]). 2Present address: The Nature Conservancy, P.O. Box 1176, Saratoga, WY 82331, USA.

Can. J. Fish. Aquat. Sci. 62: 2399–2408 (2005) doi: 10.1139/F05-149 © 2005 NRC 2400 Can. J. Fish. Aquat. Sci. Vol. 62, 2005 been directed at developing models that relate cutthroat least partially successful, and (iii) use the preferred model trout presence or abundance to habitat characteristics for electrofishing-based estimates to predict amounts and (Bozek and Rahel 1992; Kruse et al. 1997; Dunham et al. kinds of habitat necessary to sustain populations thought to 1999). In one such model, Harig and Fausch (2002) estab- meet conservation thresholds. lished that the potential for small streams to harbor self- sustaining populations of greenback cutthroat trout (Oncorhynchus clarkii stomias) and cutthroat Methods trout (Oncorhynchus clarkii virginalis) in Colorado and New was a function of summer water temperature, Field sites number of deep pools, and pool bankfull width. They con- To address the first objective, we used three sets of data: cluded that this model could also be used to assess risk of electrofishing-based abundance estimates of greenback cut- extinction of existing populations of these and closely re- throat trout in the South Platte basin in Colorado (N = lated subspecies in comparable environments. 9 streams) (Young and Guenther-Gloss 2004) and of Colo- rado River cutthroat trout in headwater streams in the Little The cutthroat trout (Oncorhynchus clarkii basin in (N = 12) and the upper Colo- pleuriticus ) is among the suite of cutthroat trout subspecies rado River basin in Colorado (N = 10) (Fig. 1). To fulfill the in the and occupies cold, high-elevation second objective, we used the ocular counts of abundance of streams in much of its remaining range (Young et al. 1996). 21 populations of greenback cutthroat trout in the South Marked declines in its distribution relative to its historical Platte and basins in Colorado and of Rio range have led to special designation by government agen- Grande cutthroat trout in the Rio Grande basin in Colorado cies (CRCT Task Force 2001) and to a petition for listing and (Harig and Fausch 2002). All populations under the US Act (US Fish and Wildlife of Colorado River cutthroat trout in Wyoming were believed Service 2004). There have been simulations of population to be indigenous, whereas those in Colorado were either in- viability under a variety of habitat structure and reintroduc- digenous or had persisted for decades after stocking (Young tion scenarios (Hilderbrand 2002, 2003) and thorough inven- et al. 1996). All Rio Grande cutthroat trout and most green- tories of sites for introductions (CRCT Task Force 2001). back cutthroat trout populations were relatively recent intro- For these reasons, the Harig and Fausch (2002) model (here- ductions (Harig and Fausch 2002; Young et al. 2002). Stocks after the translocation model) merits consideration by biolo- represented a mixed heritage; some were believed to be ge- gists working with this subspecies. In preliminary evaluations, netically pure and others were introgressed with nonindig- Harig et al. (2000) used it to assess the likely success of a enous cutthroat trout or (Oncorhynchus mykiss). reintroduced population of Colorado River cutthroat trout, Segments of named streams (from US Geological Survey and Young and Guenther-Gloss (2004) demonstrated that 1 : 24 000 maps) with allopatric populations of cutthroat model output was related to electrofishing-derived abun- trout were typically treated as the sampling units. Barriers to dance estimates of greenback cutthroat trout populations in migration (desiccated reaches, artificial barriers, or water- northern Colorado. falls) isolated most populations from upstream invasions of Initially, we assembled data on Colorado River cutthroat nonnative fishes. In Wyoming, two-way was trout populations in Colorado and Wyoming to perform sim- possible between all and the mainstem North Fork ilar analyses, but these data had two crucial distinctions Little Snake River (the lower boundary of which was from those used to develop the original model. First, most of blocked from upstream fish migration by a weir) except the the populations that we sampled had persisted indefinitely, upper portions of Deadman, Harrison, Ted, and Third creeks whereas some streams in which populations failed to be- and the main stem itself, in which water diversions isolated come established were used to develop the translocation additional populations upstream. Downstream and upstream model. Second, trout abundance was treated as a categorical segments in Deadman and Harrison creeks and the main variable (i.e., levels of translocation success) in the model, stem were treated as different sampling units, whereas the whereas we chose to examine the relation of habitat to abun- short downstream segments of Ted and Third creeks dance analyzed as a continuous variable. This also permitted (<450 m) were pooled with the main stem. In the upper us to include similarly obtained data on greenback cutthroat North Fork Little Snake main stem and in some Colorado trout from northern Colorado (Young and Guenther-Gloss streams, small (<2.5 m bankfull width), short (generally 2004). Consequently, we broadened our goals to include ex- <1000 m) named tributaries to the primary sampling streams amining relations between stream characteristics and cut- were also sampled and pooled with the primary streams for throat trout abundance, devising a method to estimate the our analyses. abundance of cutthroat trout in unsampled streams, and pro- There was broad overlap in the ranges of the physical char- viding guidance for selecting streams likely to produce pop- acteristics of these streams (Table 1) and in climate, hydrology, ulations exceeding standards for viability. Our objectives and trout population characteristics. Most were high- were to (i) develop a model relating the same habitat vari- elevation, perennial streams with snowmelt-dominated flows ables in the translocation model, plus occupied stream and gravel–cobble–rubble channels. Riparian overstory vege- length, to the abundance of age-1 and older Colorado River tation usually consisted of coniferous forests of Engelmann cutthroat trout and greenback cutthroat trout sampled by spruce (Picea engelmannii), subalpine fir (Abies lasiocarpa), electrofishing, (ii) attempt to validate the model by conduct- and lodgepole pine (Pinus contorta) interspersed with sites ing a similar analysis of the data of Harig and Fausch (2002) dominated by quaking aspen (Populus tremuloides), decidu- for greenback cutthroat trout and Rio Grande cutthroat trout ous shrubs, and meadows. Because most streams contained but limited to those streams in which translocations were at barriers, cutthroat trout probably exhibited resident life histo-

© 2005 NRC Canada Young et al. 2401

Fig. 1. Map of the US central Rocky Mountains showing 22 streams sampled to estimate cutthroat trout (Oncorhynchus clarkii)abun- dance, general locations of previous sampling, and historical ranges of Colorado River cutthroat trout (Oncorhynchus clarkii pleuriticus), greenback cutthroat trout (Oncorhynchus clarkii stomias), and Rio Grande cutthroat trout (Oncorhynchus clarkii virginalis). Circled areas denote locations of streams sampled in this study (refer to Table 1 for stream names associated with num- bers). A, streams with greenback cutthroat trout sampled by Young and Guenther-Gloss (2004) and Harig and Fausch (2002) in the South basin; B, streams with greenback cutthroat trout sampled by Harig and Fausch (2002) in the Arkansas River basin; C, streams with Rio Grande cutthroat trout sampled by Harig and Fausch (2002).

ries with movements limited to a few kilometres (Young likely to support many fish (i.e., it was <0.5 m wetted width, 1996). Cutthroat trout spawned after peak flows from late above a waterfall, or originated from a spring). All sampling May to early July and fry typically emerged from August to was conducted during late-summer low flows. September. Stream temperatures generally declined markedly We used double sampling (Cochran 1977) to estimate by late September, with ice and snow cover often beginning abundance of cutthroat trout ≥75 mm total length (age 1 and to form in October. The short growing season produced slow- older). Crews, usually consisting of one person electrofish- growing fish that rarely exceeded 250 mm total length. ing and two people netting, first made a single electrofishing pass without block nets in 25-m reaches at systematic inter- Study design vals. Intervals between electrofishing reaches varied from 50 Methods for estimating fish abundance followed Young to 400 m to ensure an adequate number of samples in the and Guenther-Gloss (2004). We measured thalweg length of shortest streams and resulted in sampling 5%–45% of each each stream with a drag tape, flagged 100-m intervals and stream (Table 1). For second-stage samples, about every electrofishing sites, and noted whether fish were present. We fourth sample reach was electrofished two or three times to surveyed at least 500 m upstream of the last observation of produce a multiple-pass removal estimate of trout abun- fish or until a stream became largely uninhabitable or un- dance. This sampling frequency ensured that at least four or

© 2005 NRC Canada 2402 Can. J. Fish. Aquat. Sci. Vol. 62, 2005

Table 1. Characteristics of cutthroat trout (Oncorhynchus clarkii) streams in Wyoming and Colorado used in the models.

%of Trout Occupied Mean pool stream abundance Elevation stream length Pools ≥30 cm bankfull Mean July water Stream sampled (≥75 mm) (m) (m) residual depth width (m) temperature (°C) Colorado River cutthroat trout (Oncorhynchus clarkii pleuriticus), Little Snake River Basin, Wyoming 1. NFLSR (lower) 10 4776 2252 14 545 385 7.5 9.7 2. Solomon 10 1705 2327 5 700 62 3.4 10.0 3. Green Timber 25 684 2523 3 150 83 3.3 9.3 4. Harrison (lower) 17 574 2518 2 557 66 2.8 9.7 5. NFLSR (upper) 17 378 2727 2 417 129 2.7 9.2 6. Rose 17 343 2365 1 670 33 3.5 10.1 7. Deadman (lower) 25 308 2584 1 322 45 4.4 9.6 8. Deadman (upper) 19 244 2740 3 175 116 3.4 8.0 9. Rhodine 23 177 2762 1 318 53 2.4 8.7 10. Ted (upper) 25 158 2724 1 940 53 2.7 8.5 11. Third (upper) 23 102 2720 1 315 45 3.0 10.3 12. Harrison (upper) 45 76 2760 553 12 2.0 8.5 Colorado River cutthroat trout), upper Colorado River Basin, Colorado 13. Buchanan 15 2531 2774 5 725 63 6.5 10.8 14. Cabin 10 1354 2914 6 054 212 5.0 9.7 15. Hamilton 23 798 2865 3 225 46 2.1 8.5 16. Trail 10 594 2780 6 185 131 3.2 9.8 17. McQueary 21 379 3182 2 671 52 5.0 7.7 18. Roaring Fork 23 278 2731 4 164 160 4.9 9.0 19. Little Vasquez 18 115 2920 2 675 130 2.9 7.6 20. Jim 16 102 2853 2 225 10 3.9 7.2 21. Middle Fork Ranch 20 54 2896 2 395 28 4.8 7.7 22. South Fork Ranch 17 30 2939 2 880 26 3.0 8.2 Greenback cutthroat trout (Oncorhynchus clarkii stomias), Basin, Colorado 5–21 170–7347 2323–2926 1 640–13 201 26–172 2.3–4.1 6.1–14.2 Greenback cutthroat trout and Rio Grande cutthroat trout (Oncorhynchus clarkii virginalis), Colorado and New Mexico 100 16–1278 2400–3293 1 000–20 500 12–361 2.3–5.4 6.0–14.6 Note: Numbers associated with stream names refer to the numbers in Fig. 1. Ranges for data on greenback cutthroat trout and Rio Grande cutthroat trout are from Harig and Fausch (2002) and Young and Guenther-Gloss (2004). NFLSR, North Fork Little Snake River. Percentage of stream sampled re- fers to the proportion of occupied stream length sampled for fish by electrofishing or visual methods. five reaches would be sampled with multiple electrofishing a separate estimator of catchability to calculate abundance of passes in each stream in case calculating stream-specific Colorado River cutthroat trout in the North Fork Little catchability was required. In wider streams, two electro- Snake River main stem in Wyoming because it was much fishing crews worked side by side. To estimate population larger and catchability was lower than for other streams in size of greenback cutthroat trout in Colorado and in tributar- this area. All estimates for greenback cutthroat trout streams ies containing Colorado River cutthroat trout in Wyoming, were based on samples collected in 1999, whereas those for we multiplied the inverse of overall catchability for each set Colorado River cutthroat trout represented the mean of four of streams by mean abundance of fish in single-pass catches annual estimates from 1996 to 1999 (Wyoming) or one-time in the sample reaches of a particular stream, extrapolated to sampling between 1998 and 2002 (Colorado). We regard the length of that stream (Bohlin et al. 1989). We deemed these population estimates as conservative because depletion overall catchability appropriate because many of the same electrofishing tends to underestimate abundance (Bohlin individuals (and the same crew leaders) formed these crews 1982; Peterson et al. 2004), although cutthroat trout in small annually and they used the same equipment to sample these streams typically yield high catchabilities (Dunham et al. waters, and within each river basin, geology, habitat , 1999; Young and Guenther-Gloss 2004). and riparian vegetation were comparable. Furthermore, we Habitat inventories varied by state. For Colorado streams, considered the stream-specific estimates based on fewer crews measured habitat in accordance with the methods of samples to be less likely to produce accurate estimates of Harig and Fausch (2002). They recorded bankfull width, population size in the set of comparable streams than would maximum depth, and tail crest depth for pools where pool an estimate of overall catchability. In contrast, the abun- length was at least half of bankfull width and censused the dance estimates for Colorado River cutthroat trout in Colo- number of pools with residual depths of at least 30 cm. For rado were not collected in the same year and stream-to- streams in Wyoming, habitat components were only mea- stream variation in catchability was higher (M.K. Young, sured in electrofishing reaches (10%–45% of each stream); unpublished data); consequently, we used stream-specific the means for reaches in a stream were extrapolated to the catchability estimates to calculate abundance. We also used rest of that stream. We regarded such sampling as adequate

© 2005 NRC Canada Young et al. 2403 because it greatly exceeded reach lengths (20–40 wetted or heteroscedasticity; visual counts required no such transfor- bankfull widths) generally used to describe geomorphic vari- mation. We selected a preferred model for each data set ables (Kaufmann et al. 1999). Also, only wetted width was based on three criteria: weights (w) of Akaike’s Information measured, so we estimated pool bankfull width for Wyo- Criterion adjusted for small sample sizes (Burnham and An- ming streams by multiplying pool wetted width by 1.5, the derson 2002), all regression coefficients significantly differ- ratio between bankfull and baseflow wetted width based on ent from zero, and ease of measurement of model variables. measurements previously taken in two of the study streams We considered all models with values of w within 10% of (P.M. Guenther-Gloss, unpublished data). the best approximating model to have reasonable levels of Water temperatures were measured with thermographs support. To correct for multiple estimates of the significance (recording interval ≤2 h) set in relatively deep, well-mixed of regression coefficients, we used a Bonferroni-corrected pools in Wyoming streams in 1997, 1998, and 2000 and in value of P ≤ 0.0125 (=0.05/4, where 4 is the number of vari- Colorado streams in 2001 and 2002. We used within- and ables for which we generated estimates). Variables in order among-stream and climatic comparisons from this 5-year pe- of their ease of measurement were water temperature = riod to develop a standardized mean July water temperature stream length → pool count = pool bankfull width. We for each stream. Full-month mean July water temperatures deemed water temperature as one of the two easiest vari- were estimated for some Wyoming streams from the ratio of ables to measure because it could be obtained from only two partial- to full-month means for nearby streams with com- site visits, to set and retrieve thermographs, although mea- plete July data. In Steelman Creek (Colorado), mean July surements needed to be collected in different years to esti- stream temperature was estimated for 2001 by subtracting mate mean values. Occupied stream length was considered the mean difference for all other sampled Colorado streams comparably easy to obtain because although the entire occu- (1.1 °C) from the 2002 temperature for Steelman Creek. We pied stream channel needed to be traversed (to measure addressed the potential effect of climatic variation between channel length), detailed surveys were not necessary and a years by examining annual deviation from the 92-year mean single visit would be adequate to estimate occupied habitat July air temperature at Steamboat Springs, Colorado (located barring dramatic environmental changes. In contrast, counts in similar montane topography approximately equidistant of pools ≥30 cm residual depth and pool bankfull width between Colorado and Wyoming field sites; Western Re- could be derived only from whole-stream inventories or ex- gional Climate Center 2003). July air temperature in 1997 trapolated estimates of measured habitat types. Because did not differ from the 92-year mean; all following years streams with electrofishing-based abundance estimates origi- were warmer than average, with 2002 the warmest of all nated from three distinct locations (the South Platte, Colo- (3.3 °C above the mean). Unadjusted mean July stream tem- rado, and Little Snake River basins), we included location as peratures also followed that pattern: all 10 Colorado streams a categorical variable in the final regression model to deter- were warmer in 2002 than in 2001, and 9 of 10 comparable mine whether there was a significant location-level effect Wyoming streams were warmer in 2000 than in 1997 or (Dunham and Vinyard 1997). We conducted additional vali- 1998. Thus, stream temperatures from 1998 to 2002 were dation of the electrofishing-based model by using a jackknife adjusted downward using the ratio of study year to 92-year approach, i.e., excluding one observation, reconstructing the mean July air temperature to minimize effects of annual cli- model with the 30 remaining observations, predicting the matic variation on water temperatures used in the model. response of the excluded observation, and examining the Finally, temperatures from multiple recording locations and correlation between predicted and observed abundances multiple sampling years were averaged (in part to address (Olden and Jackson 2000). other variables influencing water temperature, such as dis- Finally, we calculated means and 95% inverse prediction charge) to provide a single temperature value for each intervals (Zar 1984) of stream length predicted to support stream. various population sizes of age-1 and older cutthroat trout based on the preferred model for electrofishing data. This Analyses enabled direct comparisons with previous work linking trout We treated fish abundance as a continuous dependent vari- abundance to stream size (Hilderbrand and Kershner 2000). able and assessed its relation to occupied stream length (m) We did not consider predictions appropriate for the preferred and the three variables in the translocation model: mean July model for visual counts because such counts may produce water temperature (°C), pool bankfull width (m), and num- unreliable estimates of fish abundance (Bozek and Rahel ber of pools with residual depths ≥30 cm. We included the 1991; Young and Guenther-Gloss 2004). Unless otherwise latter because it was the simplest estimator of habitat avail- stated, results of statistical tests were considered significant ability and a necessary adjunct to the estimates of population if P ≤ 0.05. size. Before developing models, we created a correlation matrix to assess collinearity among the physical variables Results and avoided simultaneous inclusion of those variables with significant correlations >0.60. We then used multiple linear Because occupied stream length and number of deep regression to create models of all remaining possible subsets pools were highly correlated (r = 0.82, P < 0.001 for electro- of the four explanatory variables (N = 11 possible models; fishing estimates; r = 0.80, P < 0.001 for visual counts), we see below). We examined residuals to detect outliers and did not include both variables in any model. Of the 11 re- heteroscedasticity and to evaluate overall model fit. Conse- maining models for streams with electrofishing estimates, quently, we applied a square root transformation to three models including combinations of stream length, water electrofishing-derived abundance estimates to correct for temperature, and pool bankfull width had the highest model

© 2005 NRC Canada 2404 Can. J. Fish. Aquat. Sci. Vol. 62, 2005

Table 2. Model selection statistics associated with the four highest ranking models relating mean July water temperature, pool bankfull width, counts of pools ≥30 cm residual depth, and occupied stream length to estimates of fish abundance. ∆ Variable(s) ln(L) K AICc w Electrofishing counts, Colorado River cutthroat trout (Oncorhynchus clarkii pleuriticus) and green- back cutthroat trout (Oncorhynchus clarkii stomias) Temperature, stream length –60.926 4 131.390 0.000 0.612 Stream length –63.499 3 133.887 2.497 0.175 Temperature, pool width, stream length –60.882 5 134.165 2.775 0.153 Visual counts, greenback cutthroat trout and Rio Grande cutthroat trout (Oncorhynchus clarkii virginalis) Temperature, pool count –103.150 4 216.800 0.000 0.794 Temperature, pool width, pool count –103.002 5 220.003 3.203 0.160

Note: Models are ordered by w. ln(L), maximized log-likelihood; K, number of parameters; AICc , Akaike’s Infor- ∆ mation Criterion adjusted for small sample size; , change in AICc ; w, Akaike weight.

Table 3. Coefficients of variables of the highest ranking models relating mean July water temperature, pool bankfull width, sum of pools ≥30 cm residual depth, and occupied stream length to estimates of fish abundance. Mean July water Mean pool Pools ≥30 cm Occupied stream Constant temperature (°C) bankfull width (m) residual depth length (m) Electrofishing counts, Colorado River cutthroat trout (Oncorhynchus clarkii pleuriticus) and green- back cutthroat trout (Oncorhynchus clarkii stomias) –13.58 2.26* 0.00464* 5.15* 0.00508* –12.03 2.20* –0.37 0.00472* Visual counts, greenback cutthroat trout and Rio Grande cutthroat trout (Oncorhynchus clarkii virginalis) –820.69* 91.39* 2.32* –903.97* 92.54* 20.74 2.30* Note: Models are ordered as in Table 2. An asterisk denotes that a coefficient is statistically different from zero at P ≤ 0.0125.

Fig. 2. Relation of abundances of Colorado River cutthroat trout years. Despite that it did not have the highest model weight, (Oncorhynchus clarkii pleuriticus) in Colorado (squares) and we regarded the model with stream length alone (Fig. 2) as Wyoming (circles) and greenback cutthroat trout (Oncorhynchus the preferred model because it was the most parsimonious, clarkii stomias) in Colorado (triangles) to occupied stream required the least effort for data collection, and accounted length. The line represents the regression model (population for a large proportion of the variation in trout abundance size)1/2 = 0.00508(occupied stream length (m)) + 5.148. (r2 = 0.81). Also, differences in predicted abundances be- tween the preferred model and best approximating model were small (median = 81 fish, range = 3–789 fish). For streams with visual counts, the best approximating model in- cluded stream temperature and number of deep pools as variables. The only other model with a reasonable level of support included those two variables and pool bankfull width. Because the latter variable was again not significantly different from zero, we regarded the best approximating model as the preferred model for visual counts. Dummy regression with stream location and occupied stream length in a model using electrofishing estimates indi- cated that stream location did not substantially affect the model (i.e., neither the location coefficient (P = 0.43) nor lo- cation × length coefficient (P = 0.95) was significantly dif- weights (Table 2). Pool bankfull width was not significantly ferent from zero); thus, pooling locations in the preferred different from zero (Table 3), so the model containing this model was acceptable (Fig. 2). Jackknife validation sug- variable was not considered further. The model including gested that the preferred model for electrofishing-based data temperature and stream length had the highest model weight was very precise (r = 0.88) but biased (slope coefficient = but required more intensive data collection from several 0.80, 95% confidence interval = 0.63–0.97).

© 2005 NRC Canada Young et al. 2405

Table 4. Means and 95% inverse prediction intervals of stream length predicted to support particular population sizes of age-1 and older cutthroat trout (Oncorhynchus clarkii) and stream length predictions from Hilderbrand and Kershner (2000) assuming a density of 200 fish·km–1. 95% prediction interval Population Predicted stream Lower Upper Predicted stream length size length (m) bound bound (m) (200 fish·km–1) 500 3 387 2 923 3 800 2 500 1000 5 209 4 783 5 710 5 000 2000 7 787 7 107 8 718 10 000 2500 8 825 8 008 9 965 12 500 5000 12 901 11 497 14 907 25 000

Projections of stream length needed to produce popula- and detectability by observers (Thurow 1994). Second, effi- tions of 500–5000 age-1 and older cutthroat trout reflected ciency of visual sampling probably declined as stream size the form of the relation between abundance and occupied and depth increased (Heggenes et al. 1990), and Young and stream length (Table 4). Assuming a uniform density of Guenther-Gloss (2004) noted that visual counts ranged from 200 fish·km–1 (as in Hilderbrand and Kershner 2000) slightly 1% to 24% of electrofishing estimates and were only weakly underestimated necessary stream lengths for small popula- correlated with them. Third, the absence of deep pool counts tions but greatly overestimated habitat needed to produce the from the preferred model for electrofishing data implied that largest populations. deep pools accounted for habitat availability less well than a more generic measure of available habitat or that rules for Discussion defining deep pools (i.e., that they be at least half as long as the bankfull width and at least 30 cm residual depth) ex- In the streams that we evaluated, occupied stream length cluded much suitable habitat. Alternatively, this variable was the single best predictor of cutthroat trout abundance. may have been important in the original translocation model Bradford et al. (1997) reported a similar pattern for coho because trout are more easily observed in deep pools rather (Oncorhynchus kisutch) smolts in western North than in more turbulent or shallower sites, a bias much less America. Other studies have demonstrated that basin area, evident in electrofishing data (Heggenes et al. 1990). Fourth, arguably a surrogate of stream length, is highly correlated Bradford et al. (1997) noted that streams must be fully with the presence or size of salmonid populations (Kruse seeded for the habitat size – fish abundance relation to hold. 1998; Dunham and Rieman 1999; Dunham et al. 2002). Most streams that we sampled had supported cutthroat trout Harig and Fausch (2002) noted that basin area was the only populations for decades and had little mortality; landscape-scale variable correlated with translocation suc- thus, populations may have approached the carrying capacity cess of cutthroat trout but offered little predictive power. Oc- of the habitat, whereas Harig and Fausch (2002) examined cupied stream length, however, is a more accurate measure populations of more recent origin in which this relation of habitat availability than basin area (Dunham et al. 2002) would be less certain. Also, streams with marginally suc- because it accounts for the presence of natural barriers that cessful translocations could be markedly influenced by small constrain distributions of fish populations. Such barriers differences in pool numbers, width, or temperature, and commonly limited the upstream and downstream extent of some unsuitable patches may contain populations whose the allopatric populations that we sampled and defined habi- long-term persistence is unlikely (Morita and Yamamoto tat patches supporting discrete populations. Not addressed 2002). Consequently, we cannot regard the model based on by our models were possible barriers within the occupied visual counts as validating the model based on electrofishing patches. Whether these were current obstructions to upstream estimates but believe that this discrepancy arose primarily migration could not be ascertained because allopatric popu- from differences in the sampling methodologies used and lations of cutthroat trout occupied the areas upstream and streams chosen rather than shortcomings of the downstream. Within-patch barriers would create sub- electrofishing-based model. populations in a stream, with the upstreammost segment iso- The failure of pool bankfull width to be included in either lated from patches downstream and at greater risk of preferred model or to be significantly related to fish abun- extinction from extreme disturbance (Dunham et al. 1999; dance when considered alone, despite that longer and pre- Kruse et al. 1997). Data on emigration and immigration sumably wider streams contained greater densities of fish, rates in a more advanced model would enable these streams may have resulted from where bankfull width was measured. to be evaluated as a series of partially isolated segments Pool bankfull width is a function not only of stream size but rather than as the cumulative total of occupied habitat. also of pool type, and because plunge, dammed, lateral We attribute the contrast in the response of fish abundance scour, and trench pools have unique morphologies, variation in the different models, as a function of the square of occu- attributable to pool type could overwhelm trends in width pied stream length in the electrofishing-based model and as associated with stream size. Traditionally, measurements of a linear function of water temperature and deep pool counts bankfull width have been obtained in riffles or other channel in the model for visual counts, to four factors. First, we sus- units of fairly homogeneous width (Rosgen 1996), and we pect that colder streams probably diminished trout activity suspect that more consistent measures of overall stream

© 2005 NRC Canada 2406 Can. J. Fish. Aquat. Sci. Vol. 62, 2005 width that enabled estimates of stream area may have been cutthroat trout; four of the eight largest trout that we included in a preferred model and improved its performance. sampled overall were in the vicinity of the . Finally, we Alternatively, because abundance increased in proportion suggest caution in applying the model to other subspecies. to the square of occupied stream length in the electrofishing- For example, whereas occupied stream length was signifi- based model, this relation may reflect increases in both habi- cantly positively correlated (r2 = 0.71, P < 0.001, N = 19) tat availability and diversity. Not only is supplemental habi- with the square root of abundance of allopatric Yellowstone tat present in longer streams, but they are more likely to cutthroat trout (Oncorhynchus clarkii bouvieri) in streams in contain a complete suite of complementary habitats (e.g., the Bighorn River basin in Wyoming (Kruse et al. 2001, temperature refugia, off-channel ponds, overwintering pools, p. 6), the slope coefficient (b = 0.00259) was about half that summer rearing sites, or spawning areas suitable at varying (b = 0.00508) in the equation for streams in the central flows) (Rieman and McIntyre 1995; Schlosser and Angermeier Rocky Mountains. That abundance still increased as the 1995). The lack of upstream–downstream trends in cutthroat square of occupied stream length suggests that this func- trout density in these and other Rocky Mountain streams tional relation may have some generality and merits testing (Young and Guenther-Gloss 2004; M.K. Young, unpublished with other subspecies in small, cold streams. data) implies that habitat complexity contributes to the func- A variety of sources have hypothesized that persistence of tional response between abundance and stream length. Also, trout populations, particularly with respect to long-term evo- temporal variation in habitat availability and the magnitude lutionary potential, is related to abundances of 500–5000 of disturbance are more pronounced in smaller streams and fish (McIntyre and Rieman 1995; Allendorf et al. 1997; may influence this pattern. Hilderbrand and Kershner 2000). Our estimates of mean The square root transformation of electrofishing estimates stream lengths necessary to produce populations of these of fish abundance was necessary because the variance about sizes differed from those of Hilderbrand and Kershner the estimates increased with increasing stream size, produc- (2000) because they assumed constant cutthroat trout densi- ing a wedge-shaped distribution of points (Terrell et al. ties of 100–300 fish·km–1, whereas we found that trout den- 1996). Theoretically, quantile regression on the uppermost sity was positively correlated with occupied stream length distribution of points might suggest maximum carrying ca- and ranged from 10 to 557 fish·km–1. Nevertheless, both pacities that could be produced by streams of a given length analyses imply that many streams containing cutthroat trout (Dunham et al. 2002). Yet because none of the other habitat support populations well below thresholds thought to afford variables that we examined were included in the final regres- resilience to environmental perturbation and loss of genetic sion model, we can offer little guidance on what habitat ele- variation (Allendorf et al. 1997; McElhany et al. 2000). For ments might be suppressing fish abundance in streams below example, mean length of all recovery streams for greenback such a regression line. This variation may also have arisen in cutthroat trout was 4790 m (Young et al. 2002), for which part from the collection of abundance data in different years. we would predict a population size of 870 fish ≥75 mm Regardless, it seems more reasonable for rare subspecies of (95% confidence interval = 701–1057 fish). Similarly, 40% cutthroat trout to estimate average, rather than maximum, of streams constituting the core conservation populations for fish abundances when evaluating streams for introductions Colorado River cutthroat trout in Colorado, Utah, and Wyo- or the viability of existing populations. ming are no more than 3000 m long (US Fish and Wildlife To address issues of model precision and generality Service 2004); on average streams of this length would be (Fausch et al. 1988), we tried to ensure that streams from predicted to contain 416 or fewer fish (95% confidence different basins shared similar environmental characteristics interval = 298–554 fish). Encouragingly, the preferred model and that sample sizes were adequate. Although cross- demonstrated that relatively minor increases in habitat length validation of the electrofishing-based model demonstrates led to disproportionately greater increases in abundance, im- that it is fairly accurate, the inherent bias (probably intro- plying that extending available habitat, particularly the duced by the square root transformation of abundance) limits downstream end, might be a key strategy for enhancing pop- its generality. Therefore, applications of the preferred model ulation viability. for electrofishing estimates should be restricted to systems Selecting streams for reintroductions and evaluating via- comparable to those that we evaluated: small, high-elevation bility of existing populations of rare subspecies of cutthroat streams warm enough to permit recruitment. This model trout are politically controversial, biologically critical, and does not apply where high water temperatures (>22 °C) limit fraught with uncertainty (Young and Harig 2001). We be- survival and growth (Dunham et al. 2003; Schrank et al. lieve that use of the preferred model for electrofishing data 2003). The streams used to develop the preferred model also can reduce the ambiguity associated with predicting poten- were relatively free of human-associated disturbance, other tial average size of reintroduced populations or mean size of than summer water withdrawals in some systems. Land man- unsampled populations, with two additional caveats. First, agement has often been associated with the decline in or ab- although map-derived estimates of potential occupied stream sence of salmonid populations (Poff and Huryn 1998; length may be a useful first approximation, the importance Dunham and Rieman 1999; Pess et al. 2002); thus, popula- of barriers in structuring populations mandates that lengths tions in altered systems may not conform to model predic- be verified by field surveys. Second, there is substantial sup- tions. In addition, we urge care in applying the model to port that mean July water temperature was positively related streams subsidized by lake populations or interrupted by to fish abundance in models based on electrofishing esti- large numbers of beaver ponds. Although we included one mates and visual counts, probably reflecting increases in such stream, McQueary Creek, in our sampling, a headwater productivity at warmer temperatures. More importantly, be- lake possibly inflated the abundance and certainly the size of cause low water temperatures (mean summer water tempera-

© 2005 NRC Canada Young et al. 2407 ture <7.5 °C) can inhibit cutthroat trout recruitment (Harig in the states of Colorado, Utah, and Wyoming. Colorado Divi- and Fausch 2002; Peterson 2002; M. Coleman and K.D. sion of Wildlife, Fort Collins, Co. Fausch, Department of Fishery and Wildlife Biology, Colo- Duff, D. (Editor). 1996. Conservation assessment for inland cutthroat rado State University, Fort Collins, CO 80523, USA, unpub- trout: distribution, status and habitat management implications. US lished data), prudent management dictates that decisions on For. Serv. Northern, Rocky Mountain, Intermountain, and South- conservation priorities and habitat protection must also con- western Regions, Logan, Utah. sider whether summer water temperatures surpass the Dunham, J.B., and Rieman, B.E. 1999. Metapopulation structure of threshold necessary for successful recruitment. A single year : influences of physical, biotic, and geometrical land- of data collection might be sufficient to demonstrate whether scape characteristics. Ecol. Appl. 9: 642–655. temperatures are likely to be sufficiently warm. Finally, in Dunham, J.B., and Vinyard, G.L. 1997. Incorporating stream level the event that estimates of mean July water temperature and variability into analyses of site level fish habitat relationships: stream length are available, use of the model containing both some cautionary examples. Trans. Am. Fish. Soc. 126: 323–329. variables will provide slightly more accurate estimates of Dunham, J.B., Peacock, M.M., Rieman, B.E., Schroeter, R.E., and population size. Vinyard, G.L. 1999. Local and geographic variability in the dis- tribution of stream-living . Trans. Am. Fish. Soc. 128: 875–889. Acknowledgements Dunham, J.B., Cade, B.S., and Terrell, J.W. 2002. Influences of spatial and temporal variation on fish–habitat relationships de- We thank many individuals for help in the field and G. fined by regression quantiles. Trans. Am. Fish. Soc. 131: 86–98. Horstman, K. Rogers, and B. Atkinson (Colorado Division Dunham, J.B., Schroeter, R., and Rieman, B. 2003. Influence of of Wildlife) and A. Harig () for sharing data. maximum water temperature on occurrence of Lahontan cutthroat We also thank J. Dunham, A. Harig, D. Schmetterling, K. trout within streams. N. Am. J. Fish. Manag. 23: 1042–1049. Rogers, and an anonymous reviewer for their comments on Fausch, K.D., Hawkes, C.L., and Parsons, M.G. 1988. Models that earlier drafts of the manuscript and R. King for a thorough predict standing crop of stream fish from habitat variables: biometrical review. The US Forest Service through the Arap- 1950–1985. US For. Serv. Gen. Tech. Rep. PNW-GTR-213. aho-Roosevelt National Forests, Medicine Bow-Routt Na- Harig, A.L., and Fausch, K.D. 2002. Minimum habitat require- tional Forests, Rocky Mountain Research Station, Region 2 ments for establishing translocated cutthroat trout populations. Fish Habitat Relationships Program, and the interregional Ecol. Appl. 12: 535–551. Inland Cutthroat Trout Initiative provided funding for this Harig, A.L., Fausch, K.D., and Guenther-Gloss, P.M. 2000. Appli- work. cation of a model to predict success of cutthroat trout translocations in central and southern Rocky Mountain streams. In Wild trout VII. Edited by D. Schill, S. Moore, P. Byorth, and References B. Hamre. Trout Unlimited, Bethesda, Md. pp. 141–148. Allendorf, F.W., Bayles, D., Bottom, D.L., Currens, K.P., Frissell, Heggenes, J., Brabrand, Å., and Saltveit, S.J. 1990. Comparison of C.A., Hankin, D., Lichatowich, J.A., Nehlsen, W., Trotter, P.C., three methods for studies of stream habitat use by young brown and Williams, T.H. 1997. Prioritizing Pacific salmon stocks for trout and . Trans. Am. Fish. Soc. 119: 101–111. conservation. Conserv. Biol. 11: 140–152. Hepworth, D.K., Ottenbacher, M.J., and Berg, L.N. 1997. Distribution Behnke, R.J. 1992. Native trout of western . Am. and abundance of native Bonneville cutthroat trout (Oncorhynchus Fish. Soc. Monogr. No. 6. clarki utah) in southwestern Utah. Nat. 57: 11–20. Bohlin, T. 1982. The validity of removal method for small Hilderbrand, R.H. 2002. Simulating supplementation strategies for populations — consequences for electrofishing practice. Inst. restoring and maintaining stream resident cutthroat trout popula- Freshw. Res. Drottningholm Rep. 60: 15–18. tions. N. Am. J. Fish. Manag. 22: 879–887. Bohlin, T., Hamrin, S., Heggberget, T.G., Rasmussen, G., and Hilderbrand, R.H. 2003. The roles of carrying capacity, immigra- Saltveit, S.J. 1989. Electrofishing — theory and practice with tion, and population synchrony on persistence of stream-resident special emphasis on salmonids. Hydrobiologia, 173: 9–43. cutthroat trout. Biol. Conserv. 110: 257–266. Bozek, M.A., and Rahel, F.J. 1991. Comparison of streamside vi- Hilderbrand, R.H., and Kershner, J.L. 2000. Conserving inland cut- sual counts to electrofishing estimates of Colorado River cut- throat trout in small streams: how much stream is enough? N. throat trout fry and adults. N. Am. J. Fish. Manag. 11: 38–42. Am. J. Fish. Manag. 20: 513–520. Bozek, M.A., and Rahel, F.J. 1992. Generality of microhabitat suit- Kaufmann, P.R., Levine, P., Robison, E.G., Seeliger, C., and Peck, ability models for young Colorado River cutthroat trout D.V. 1999. Quantifying physical habitat in wadeable streams. (Oncorhynchus clarki pleuriticus) across sites and among years EPA/620/R-99/003. US Environmental Protection Agency, Wash- in Wyoming streams. Can. J. Fish. Aquat. Sci. 49: 552–564. ington, D.C. Bradford, M.J., Taylor, G.C., and Allan, J.A. 1997. Empirical re- Kruse, C.G. 1998. Influence of non-native trout and geomorphol- view of smolt abundance and the prediction of ogy on the distributions of indigenous trout in the Yellowstone smolt production at the regional level. Trans. Am. Fish. Soc. River drainage of Wyoming. Ph.D. dissertation, University of 126: 49–64. Wyoming, Laramie, Wy. Burnham, K.P., and Anderson, D.R. 2002. Model selection and Kruse, C.G., Hubert, W.A., and Rahel, F.J. 1997. Geomorphic in- multimodel inference: a practical information-theoretic ap- fluences on the distribution of Yellowstone cutthroat trout in the proach. 2nd ed. Springer, New York. Absaroka Mountains, Wyoming. Trans. Am. Fish. Soc. 126: Cochran, W.G. 1977. Sampling techniques. 3rd ed. John Wiley & 418–427. Sons, New York. Kruse, C.G., Hubert, W.A., and Rahel, F.J. 2001. As assessment of CRCT Task Force. 2001. Conservation agreement and strategy for headwater isolation as a conservation strategy for cutthroat trout Colorado River cutthroat trout (Oncorhynchus clarki pleuriticus) in the Absaroka Mountains of Wyoming. Northwest Sci. 75: 1–11.

© 2005 NRC Canada 2408 Can. J. Fish. Aquat. Sci. Vol. 62, 2005

McElhany, P., Ruckelshaus, M.H., Ford, M.J., Wainwright, T.C., volving Bonneville cutthroat trout. Trans. Am. Fish. Soc. 132: and Bjorkstedt, E.P. 2000. Viable salmonid populations and the 100–109. recovery of evolutionarily significant units. NOAA Tech. Memo. Terrell, J.W., Cade, B.S., Carpenter, J., and Thompson, J.M. 1996. NMFS-NWFSC-42. Modeling stream fish habitat limitations from wedge-shaped McIntyre, J.D., and Rieman, B.E. 1995. Westslope cutthroat trout. In patterns of variation in standing stock. Trans. Am. Fish. Soc. Conservation assessment for inland cutthroat trout. Edited by M.K. 125: 104–117. Young. US For. Serv. Gen. Tech. Rep. RM-GTR-256. pp. 1–15. Thurow, R.F. 1994. Underwater methods for study of salmonids in Morita, K., and Yamamoto, S. 2002. Effects of habitat fragmenta- the Intermountain West. US For. Serv. Gen. Tech. Rep. INT- tion by damming on the persistence of stream-dwelling charr GTR-307. populations. Conserv. Biol. 16: 1318–1323. US Fish and Wildlife Service. 1998. Greenback cutthroat trout re- Olden, J.D., and Jackson, D.A. 2000. Torturing data for the sake of covery plan. US Fish and Wildlife Service, Denver, Co. generality: how valid are our regression models? Ecoscience, 7: US Fish and Wildlife Service. 2004. Endangered and threatened 501–510. wildlife and plants; 90-day finding on a petition to list the Colo- Pess, G.R., Montgomery, D.R., Steel, E.A., Bilby, R.E., Feist, B.E., rado River cutthroat trout. Fed. Regist. 69: 21151–21158. and Greenberg, H.M. 2002. Landscape characteristics, land use, Western Regional Climate Center. 2003. Monthly average air tem- and coho salmon (Oncorhynchus kisutch) abundance, Snohomish perature at Steamboat Springs, Colorado (Site 057936) for the River, Wash., U.S.A. Can. J. Fish. Aquat. Sci. 59: 613–623. period 1910–2003. Desert Research Institute. Available from Peterson, D.P. 2002. Population ecology of an invasion: demogra- http://www.wrcc.dri.edu [accessed 10 December 2003]. phy, dispersal, and effects of nonnative on native Young, M.K. 1996. Summer movements and habitat use by Colo- cutthroat trout. Ph.D. dissertation, Colorado State University, rado River cutthroat trout in small, montane streams. Can. J. Fort Collins, Co. Fish. Aquat. Sci. 53: 1403–1408. Peterson, J.T., Thurow, R.F., and Guzevich, J.W. 2004. An evalua- Young, M.K., and Guenther-Gloss, P.M. 2004. Population charac- tion of multipass electrofishing for estimating the abundance of teristics of greenback cutthroat trout in streams: their relation to stream-dwelling salmonids. Trans. Am. Fish. Soc. 133: 462–475. model predictions and recovery criteria. N. Am. J. Fish. Manag. Poff, N.L., and Huryn, A.D. 1998. Multi-scale determinants of sec- 24: 184–197. ondary production in Atlantic salmon ( salar) streams. Young, M.K., and Harig, A.L. 2001. A critique of the recovery of Can. J. Fish. Aquat. Sci. 55(Suppl.): 201–217. greenback cutthroat trout. Conserv. Biol. 15: 1575–1584. Rieman, B.E., and McIntyre, J.D. 1995. Occurrence of bull trout in Young, M.K., Schmal, R.N., Kohley, T.W., and Leonard, V.G. naturally fragmented habitat patches of varied size. Trans. Am. 1996. Conservation status of Colorado River cutthroat trout. US Fish. Soc. 124: 285–296. For. Serv. Gen. Tech. Rep. RM-GTR-282. Rosgen, D. 1996. Applied river morphology. Wildland Hydrology, Young, M.K., Harig, A.L., Rosenlund, B., and Kennedy, C. 2002. Pagosa Springs, Co. Recovery history of greenback cutthroat trout: population char- Schlosser, I.J., and Angermeier, P.L. 1995. Spatial variation in de- acteristics, hatchery involvement, and bibliography. Version 1.0. mographic processes of lotic fishes: conceptual models, empiri- US For. Serv. Gen. Tech. Rep. RMRS-GTR-88WWW. Available cal evidence, and implications for conservation. Am. Fish. Soc. from http://www.fs.fed.us/rm/pubs/rmrs_gtr88. Symp. 17: 392–401. Zar, J.H. 1984. Biostatistical analysis. 2nd ed. Prentice-Hall, Schrank, A.J., Rahel, F.J., and Johnstone, H.C. 2003. Evaluating Englewood Cliffs, N.J. laboratory-derived thermal criteria in the field: an example in-

© 2005 NRC Canada